CN110329249B - Automobile forward collision early warning control system and method of cyclic neural network - Google Patents

Automobile forward collision early warning control system and method of cyclic neural network Download PDF

Info

Publication number
CN110329249B
CN110329249B CN201910587792.XA CN201910587792A CN110329249B CN 110329249 B CN110329249 B CN 110329249B CN 201910587792 A CN201910587792 A CN 201910587792A CN 110329249 B CN110329249 B CN 110329249B
Authority
CN
China
Prior art keywords
vehicle
driver
danger
remote control
control terminal
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201910587792.XA
Other languages
Chinese (zh)
Other versions
CN110329249A (en
Inventor
吴超仲
熊盛光
贺宜
郭柏晗
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Wuhan University of Technology WUT
Original Assignee
Wuhan University of Technology WUT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Wuhan University of Technology WUT filed Critical Wuhan University of Technology WUT
Priority to CN201910587792.XA priority Critical patent/CN110329249B/en
Publication of CN110329249A publication Critical patent/CN110329249A/en
Application granted granted Critical
Publication of CN110329249B publication Critical patent/CN110329249B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W30/00—Purposes of road vehicle drive control systems not related to the control of a particular sub-unit, e.g. of systems using conjoint control of vehicle sub-units
    • B60W30/08—Active safety systems predicting or avoiding probable or impending collision or attempting to minimise its consequences
    • B60W30/09—Taking automatic action to avoid collision, e.g. braking and steering
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W30/00—Purposes of road vehicle drive control systems not related to the control of a particular sub-unit, e.g. of systems using conjoint control of vehicle sub-units
    • B60W30/08—Active safety systems predicting or avoiding probable or impending collision or attempting to minimise its consequences
    • B60W30/095—Predicting travel path or likelihood of collision
    • B60W30/0953—Predicting travel path or likelihood of collision the prediction being responsive to vehicle dynamic parameters
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
    • B60W50/08—Interaction between the driver and the control system
    • B60W50/14—Means for informing the driver, warning the driver or prompting a driver intervention
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
    • G01D21/00—Measuring or testing not otherwise provided for
    • G01D21/02—Measuring two or more variables by means not covered by a single other subclass
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S19/00—Satellite radio beacon positioning systems; Determining position, velocity or attitude using signals transmitted by such systems
    • G01S19/01—Satellite radio beacon positioning systems transmitting time-stamped messages, e.g. GPS [Global Positioning System], GLONASS [Global Orbiting Navigation Satellite System] or GALILEO
    • G01S19/13—Receivers
    • G01S19/14—Receivers specially adapted for specific applications
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G08—SIGNALLING
    • G08C—TRANSMISSION SYSTEMS FOR MEASURED VALUES, CONTROL OR SIMILAR SIGNALS
    • G08C17/00—Arrangements for transmitting signals characterised by the use of a wireless electrical link
    • G08C17/02—Arrangements for transmitting signals characterised by the use of a wireless electrical link using a radio link
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/16—Anti-collision systems
    • G08G1/166—Anti-collision systems for active traffic, e.g. moving vehicles, pedestrians, bikes
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
    • B60W50/08—Interaction between the driver and the control system
    • B60W50/14—Means for informing the driver, warning the driver or prompting a driver intervention
    • B60W2050/143—Alarm means
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W2554/00—Input parameters relating to objects
    • B60W2554/80—Spatial relation or speed relative to objects
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W2554/00—Input parameters relating to objects
    • B60W2554/80—Spatial relation or speed relative to objects
    • B60W2554/801—Lateral distance

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Mechanical Engineering (AREA)
  • Transportation (AREA)
  • Remote Sensing (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Theoretical Computer Science (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Human Computer Interaction (AREA)
  • Traffic Control Systems (AREA)
  • Regulating Braking Force (AREA)

Abstract

The invention discloses a car forward collision early warning control system and method of a recurrent neural network, and the system of the invention comprises: the device comprises a speed sensor, an acceleration sensor, a distance measuring sensor, a main controller, a brake module, an early warning prompter, a GPS module, a wireless transmission module and a remote control terminal. The remote control terminal judges whether the self vehicle and the front vehicle keep a running state and whether a forward collision danger exists or not; if the forward collision risk exists, the remote control terminal predicts the expected braking intensity of the driver through a recurrent neural network; the remote control terminal calculates the alarm time; when a forward collision danger exists, the driver does not carry out danger avoiding operation before the alarming moment, and the early warning prompter gives an alarm; if the driver carries out danger avoiding operation before the alarming moment, the remote control terminal further judges whether a forward collision danger exists; calculating the compensation braking intensity; and compensation of the braking intensity is performed. The invention effectively reduces the occurrence rate of forward collision.

Description

Automobile forward collision early warning control system and method of cyclic neural network
Technical Field
The invention relates to the field of intelligent networked automobile safety, in particular to an automobile forward collision early warning control system and method of a recurrent neural network.
Background
Traffic accidents have become a serious challenge in modern society. According to statistics, about 60 percent of various traffic accidents in China are collision accidents, and the forward collision of automobiles is the most common accident form. The traffic accident causing the vehicle forward collision is mainly caused by two aspects: on one hand, the influence of objective environmental conditions is caused, for example, under the bad weather with poor visibility such as sand wind, snow, rain, fog and the like, a driver has identification obstacles on a front vehicle, the braking process of the vehicle is difficult to accurately estimate, and a forward collision traffic accident is caused; on the other hand, the subjective factors of the driver mainly include fatigue driving, careless driving, misoperation and the like. The current automobile forward collision danger avoiding method mainly focuses on the aspects of setting road signs for reminding the keeping of the automobile distance, calculating the driving information of the front automobile through a sensor of the automobile, calculating the danger based on automobile dynamics and the like.
In the conventional method for acquiring the calculation parameters of the forward collision of the automobile, for example, application number CN201310503550, a false alarm detection method and device for the forward collision alarm acquire image signals of a front vehicle to form video data through a camera mounted in the front of the automobile, and acquire information such as the speed of the front vehicle. For example, application No. CN201610402224, a method and a system for early warning of rear-end collision prevention of an automobile, which are implemented by adding external sensors such as a GPS sensor and an acceleration sensor on the automobile, realizing information interaction between a road and the automobile through a wireless communication device and a bluetooth function, and establishing a forward collision model. For example, application No. CN201510420512, a rear-end collision early warning method and system, which is implemented by adding a GPS sensor to a vehicle to analyze longitude and latitude information of a preceding vehicle and a following vehicle in real time, so as to calculate forward collision calculation related parameters.
The existing automobile forward collision parameter acquisition and model establishment are mainly based on information obtained by an automobile sensor to infer the information of a front automobile, and a unified threshold value is set to judge the forward collision danger by analyzing the distance between automobiles and the relative speed of the automobiles. However, in actual situations, different drivers have differences in forward collision danger sensing and emergency operation, and the information of the front vehicle calculated by the vehicle sensor has a large error, so that the acquired information is not accurate enough and has high delay. Along with the development of intelligent networking automobile technology, efficient data interaction can avoid that environmental disturbance has certain "predictability" to preceding vehicle, can also carry out coupling analysis with preceding vehicle and car information, accurately acquire preceding collision correlation calculation parameter to give the driver early warning, control intervention when necessary, in order to avoid the emergence of preceding collision.
Disclosure of Invention
The invention aims to solve the problems that: how to avoid the limitation of error of a self-vehicle sensor on the calculation of the front vehicle data and realize the self-learning of the forward risk avoiding operation of a driver, and an automobile forward collision early warning control system and method of a recurrent neural network are established.
The technical scheme of the system of the invention is as follows: a car forward collision early warning control system of a recurrent neural network is characterized by comprising: the system comprises a speed sensor, an acceleration sensor, a distance measuring sensor, a main controller, a brake module, an early warning prompter, a GPS module, a wireless transmission module and a remote control terminal;
the main controller is respectively connected with the speed sensor, the acceleration sensor, the distance measuring sensor, the braking module, the early warning prompter and the wireless transmission module in sequence through leads; the wireless transmission module is connected with the remote control terminal in a wireless communication mode.
Preferably, the speed sensor is mounted on the vehicle and used for acquiring the speed of the vehicle;
preferably, the acceleration sensor is mounted on the vehicle and used for acquiring the acceleration of the vehicle;
preferably, the distance measuring sensor is mounted at the front end of the vehicle and is used for collecting the distance between the self vehicle and the front vehicle;
preferably, the early warning prompter is arranged near a driver on the vehicle, comprises an early warning indicator lamp and a buzzer and is used for prompting the driver to brake;
preferably, the braking module is mounted on a vehicle and used for avoiding forward collision and compensating the braking strength of the whole vehicle;
preferably, the GPS module is arranged on the vehicle and used for collecting the position information of the vehicle;
preferably, the main controller is mounted on a vehicle and used for collecting the vehicle speed collected by a speed sensor, the vehicle acceleration collected by an acceleration sensor, the vehicle position information collected by a GPS module and the distance between the vehicle and the front vehicle collected by a distance measurement sensor, and determining whether the early warning prompter and the brake module work or not according to the feedback result of the remote control terminal;
preferably, the wireless transmission module is mounted on a vehicle and used for transmitting vehicle information acquired by the main controller to the remote control terminal and transmitting feedback information from the remote control terminal to the main controller;
preferably, the remote control terminal is used for analyzing and receiving vehicle information, judging the position of a front vehicle according to the GPS position information, completing self-learning of the expected braking strength of the driver and judging whether a forward collision risk exists or not.
The invention discloses a method for early warning and controlling automobile forward collision of a recurrent neural network, which comprises the following steps:
step 1: the remote control terminal judges whether the self vehicle and the front vehicle keep a running state and have a forward collision danger or not according to the speed of the self vehicle, the acceleration of the self vehicle, the speed of the front vehicle, the acceleration of the front vehicle and the distance between the self vehicle and the front vehicle at the current moment;
step 2: if the forward collision risk exists, the remote control terminal expects the brake intensity a of the driver through the recurrent neural networkeCarrying out prediction;
and step 3: the remote control terminal calculates the alarm time;
and 4, step 4: when there is a forward collision danger, the driver is at warning time TaBefore, danger avoiding operation is not carried out, the remote control terminal transmits danger signals to the early warning prompter through the wireless transmission module, and the main controller enables the early warning prompter to work and gives out an alarm.
And 5: if the driver carries out danger avoiding operation before the alarming moment, the remote control terminal further judges whether a forward collision danger exists;
step 6: calculating the compensation braking intensity;
and 7: when the braking intensity compensation is needed, the remote control terminal transmits the minimum compensation braking intensity to the main controller through the wireless transmission module, the main controller sends a command to the braking module, and the braking module completes the compensation of the braking intensity through the regulation and control of the main brake and the auxiliary braking device.
Preferably, in the step 1, the remote control terminal confirms the front vehicle according to the GPS information of the vehicle in the intelligent network connection;
in step 1, the speed of the bicycle is vrThe acceleration of the bicycle is arThe speed of the front vehicle is vfThe acceleration of the front vehicle is afThe distance between the bicycle and the front bicycle is D0；
The main controller receives v acquired by the speed sensorrAnd sending the data to a remote control terminal through a wireless transmission module;
the main controller receives a acquired by the acceleration sensorrAnd sending the data to a remote control terminal through a wireless transmission module;
d acquired by main controller receiving distance measuring sensor0And sending the data to a remote control terminal through a wireless transmission module;
in the step 1, the step of judging whether the self vehicle and the front vehicle keep the driving state or not has the risk of forward collision is as follows:
if ar>afThe speed of the bicycle is gradually faster than that of the front bicycle, so that the danger of forward collision exists;
if ar<afAnd v isr<vfThe distance between the bicycle and the front bicycle is enlarged, and the danger of forward collision is avoided;
if ar<afAnd v isr>vfIf D is>0, no forward collision danger exists, and if D is less than or equal to 0, the forward collision danger exists;
d is specifically calculated as follows:
Figure BDA0002115055340000041
wherein D is when vr＝vfThe distance between the current vehicle and the preceding vehicle;
preferably, the braking strength a expected for the driver in step 2eThe specific process for predicting comprises the following steps:
step 2.1: constructing a training set of a recurrent neural network;
step 2.1, constructing a training set of the recurrent neural network as follows:
the remote control terminal extracts relevant parameters of the braking time in the latest n times of forward risk avoiding process of the self-vehicle as a training set, wherein m is the extracted 1 st forward risk avoiding process, and the training set is as follows:
xi＝({v'r,i,v'f,i,a'r,i,a'f,i,D'0,i},0＜m≤i≤n+m,m,n,i∈Z)
wherein, v'r,iIs the speed v 'of the vehicle at the braking moment in the ith forward danger avoiding process'f,iIs the front vehicle speed a 'of the braking time in the ith forward danger avoiding process'r,iIs the self vehicle acceleration at the braking moment in the ith forward danger avoiding process'f,iIs the front vehicle acceleration D 'of the braking moment in the ith forward danger avoiding process'0,iThe distance between the self vehicle and the front vehicle at the braking moment in the ith forward danger avoiding process is obtained;
recording the braking intensity set y of the braking moment in the last n times of forward risk avoidance processi：
yi＝({ad,i}，m≤i≤n+m,m,n,i∈Z)
Wherein, ad,iThe real braking strength at the braking moment in the ith forward risk avoiding process is obtained;
step 2.2: training the cyclic neural network according to the training set to obtain a trained cyclic neural network;
the specific process of training the recurrent neural network according to the training set in step 2.2 is as follows:
determining a specific structure of a recurrent neural model, and constructing a recurrent neural network model; the constructed recurrent neural model uses a model with 1 input layer, 5 hidden layers and 1 output layer.
Model initialization: randomly initializing a weight matrix U, W, V and bias matrixes b and c in the model parameters; the hidden state of the recurrent neural network model at the braking moment in the ith forward risk avoidance process is recorded as hiAnd recording the predicted value of the model
Figure BDA0002115055340000042
The activation function f (x) being generally tanh, b is a bias in a linear relationship, the activation function g (x) is typically a Softmax function; a recurrent neural network can generally be written as follows:
Figure BDA0002115055340000051
Figure BDA0002115055340000052
forward propagation training: inputting training sample data into a recurrent neural model, obtaining a predicted value of the recurrent neural model under initial model parameters through forward propagation, and continuously reducing the predicted value
Figure BDA0002115055340000053
With the true value yiAdjusting the model parameters by the difference value;
and (3) back propagation training: selecting a loss function of the model as an optimization target, and taking a model parameter weight matrix U, W, V and bias matrixes b and c as optimization objects; iterating the model parameters by using a gradient descent method according to the error;
a cross entropy function L oss was chosen as the loss function, denoted L, expressed as follows:
Figure BDA0002115055340000054
calculating the gradient of the weight matrix V and the bias matrix c:
Figure BDA0002115055340000055
Figure BDA0002115055340000056
computing i-time hidden state gradientiWhen reversely propagating, the gradient loss of i is determined by the gradient loss corresponding to the current forward collision avoidance and the gradient loss of the next forward collision avoidance i +1, and the depth spirit is referredOver a networki+1Recursion to each otheriThe function diag represents taking the diagonal elements of the matrix:
Figure BDA0002115055340000057
calculating the gradients of the weight matrix W, U and the bias matrix b:
Figure BDA0002115055340000058
Figure BDA0002115055340000059
Figure BDA00021150553400000510
through repeated iteration:
step 2.3: determining a trained recurrent neural network model;
the specific process of training the recurrent neural network according to the training set in step 2.3 is as follows:
using the same training set xiInputting the optimized recurrent neural network model again;
combining forward propagation training and backward propagation training to compare predicted values
Figure BDA0002115055340000061
With the true value yiError between, predicted brake strength
Figure BDA0002115055340000062
With true braking strength yiThe average error should be less than a certain threshold. If the error meets the requirement, determining the model parameters; if the error does not meet the requirement, the step 2.2 is repeated to adjust the parameters until the error meets the requirement;
step 2.4: the remote control terminal sends the current speed v of the vehiclerSpeed v of front vehiclefAcceleration a of bicyclerFront truck and truckSpeed afDistance D between the bicycle and the front bicycle0Substituting the optimized recurrent neural network model to continuously predict the expected brake intensity of the drivere；
Preferably, the step 3 of calculating, by the remote control terminal, the allowable risk avoidance operation time of the driver specifically includes:
the current time is T0The driver's expected braking intensity is ae；
At the latest operation time T when the driver allows danger avoidanceeThe speed of the bicycle is ve,rThe speed of the front vehicle is ve,fThe distance D (t) between the two vehicles after the driver brakes at the latest operation time allowing danger avoidance is calculated according to the following formula:
ve,r＝(Te-T0)·ar+vr
ve,f＝(Te-T0)·af+vf
Figure BDA0002115055340000063
if the driver wants to ensure that the driver does not collide with the front vehicle, the latest operation time T for avoiding danger is allowed for the drivereD (T) is equal to or greater than 0, i.e., the discriminant Δ is equal to or greater than 0, and the critical condition is that Δ is equal to 0 and T is equal toe-T0For the driver's allowable safe-keeping operation time, d (t) the discriminant Δ is calculated as follows:
Figure BDA0002115055340000064
wherein, T0Is the current time, TeLatest operation moment v of driver's permission to avoid dangere,rVehicle speed v of the vehicle at the latest operating moment allowed for the driver to avoid dangere,fVehicle speed before the latest operation time allowed for driver to avoid danger, vrFor the current time the speed of the bicycle, arAcceleration of the vehicle at the present moment, vfFor the current time the speed of the bicycle, afAcceleration of the vehicle at the present moment, D0Is the distance between the current vehicle and the preceding vehicle at the current moment, D(t) is a distance function between two vehicles after the driver brakes at the latest operation time allowed to avoid danger, and t is the time after the driver brakes at the latest operation time allowed to avoid danger;
the latest operation time T allowed by the driver to avoid danger can be calculated through the formulae；
However, in practical cases, the alarm time TaShould be earlier than the latest operation time T of the driver for avoiding dangereOn the one hand, the driver should be given an emergency response time t after the alarm is givendOn the other hand, the acceleration change of the automobile is a continuous process and needs a certain response time tvThen alarm time TaCan be expressed as follows:
Ta＝Te-td-tv
preferably, the step 5 further determines whether there is a risk of forward collision by the remote control terminal:
the remote control terminal brakes the real braking time T of the driverdWith true braking strength adCalculating the distance D (t) between two vehicles after the driver brakes at the latest operation time allowing risk avoidance:
ve,r＝(Te-Td)·a'r+v'r
ve,f＝(Te-Td)·a'f+v'f
Figure BDA0002115055340000071
the remote control terminal brakes according to the real braking time T of the driverdWith true braking strength adCalculating D (t) discriminant Δ:
Figure BDA0002115055340000072
wherein, TdFor the actual braking moment, TeLatest operation moment v of driver's permission to avoid dangere,rVehicle speed v of the vehicle at the latest operating moment allowed for the driver to avoid dangere,fAllowing avoidance for driverVehicle speed v 'of front vehicle at latest operation time'rIs the speed of the vehicle at the moment of real braking of the driver, a'rIs the acceleration v 'of the driver at the real braking moment'fIs the speed of the vehicle at the moment of real braking of the driver, a'fIs the front vehicle acceleration of the driver at the real braking moment D'0The distance between the self vehicle and the front vehicle at the real braking moment of the driver, D (t) is a function of the distance between the two vehicles after the driver brakes at the latest operation moment allowing danger avoidance, and t is the time after the driver brakes at the latest operation moment allowing danger avoidance;
if delta is less than 0, a forward collision danger exists, the remote control terminal transmits a danger signal to the main controller through the wireless transmission module, and the main controller enables the early warning controller to work and gives an alarm;
preferably, the compensation braking intensity in step 6 is calculated as:
when the alarm is given, the driver reacts at the emergency reaction time tdShould be operated accordingly, at the vehicle response time TvWhen the danger avoiding operation is completed, the remote control terminal transmits a danger removing signal to the main controller through the wireless transmission module, and the main controller stops the early warning controller to stop working and remove the alarm; otherwise, the remote control terminal calculates whether the braking intensity needs to be compensated or not through the safety distance; at the vehicle response time TvThe remote control terminal extracts the speed v of the vehiclev,rSpeed v of the preceding vehiclev,fAcceleration a of bicyclev,rAcceleration a of the preceding vehiclev,fDistance D between the bicycle and the preceding bicyclevWhen v isv,r＝vv,fAt a distance of D1：
Figure BDA0002115055340000081
Tv＝Te-tv
When D is present1When the alarm is more than or equal to 0, the remote control terminal transmits the danger removing signal to the main controller through the wireless transmission module, and the main controller enables the early warning controller to stop working and remove the alarm. When D is present1<When 0, the remote control terminal transmits information to let the main controller orderThe brake module carries out brake intensity compensation with a compensation value of acAfter compensation control, D should be satisfied1≥0：
Figure BDA0002115055340000082
From the above formula, a minimum value a of the compensated braking intensity can be derivedcminComprises the following steps:
Figure BDA0002115055340000083
wherein, TvAs the vehicle response time, TeThe latest operation time t allowed by the driver to avoid dangerdDriver emergency response time, vv,rVehicle speed, v, of the vehicle at the time of responsev,fThe vehicle speed before the time of response of the vehicle, av,rAcceleration of the vehicle from the moment of response of the vehicle, av,fAcceleration of the vehicle ahead of the time of response of the vehicle, acTo compensate for braking strength, DvDistance between the vehicle and the preceding vehicle at the time of response of the vehicle, D1Is when v isv,r＝vv,fThe distance between the bicycle and the front bicycle.
The invention has the advantages that the vehicle information is accurately acquired through intelligent network connection, the driving habits of different drivers can be adapted through the self-learning of the recurrent neural network, and the forward collision is effectively predicted and reduced.
Drawings
FIG. 1: the method is characterized by comprising the following steps of (1) providing an intelligent network automobile forward collision control system diagram;
FIG. 2: is a flow chart of the method of the present invention;
FIG. 3: is a self-learning algorithm block diagram of the recurrent neural network;
FIG. 4: a forward collision timeline diagram.
Detailed Description
In order to facilitate the understanding and practice of the present invention for those of ordinary skill in the art, the present invention will be described in further detail with reference to the accompanying drawings and examples, it being understood that the examples described herein are for purposes of illustration and explanation only and are not intended to be limiting.
As shown in fig. 1, the technical solution of the system in the embodiment of the present invention is: a car forward collision early warning control system of a recurrent neural network is characterized by comprising: the system comprises a speed sensor, an acceleration sensor, a distance measuring sensor, a main controller, a brake module, an early warning prompter, a GPS module, a wireless transmission module and a remote control terminal;
the main controller is respectively connected with the speed sensor, the acceleration sensor, the distance measuring sensor, the braking module, the early warning prompter and the wireless transmission module in sequence through leads; the wireless transmission module is connected with the remote control terminal in a wireless communication mode.
The speed sensor is arranged on the vehicle and used for acquiring the speed of the vehicle;
the acceleration sensor is arranged on the vehicle and used for acquiring the acceleration of the vehicle;
the distance measuring sensor is arranged at the front end of the vehicle and is used for collecting the distance between the self vehicle and the front vehicle;
the early warning prompter is arranged near a driver on the vehicle, comprises an early warning indicator lamp and a buzzer and is used for prompting the driver to brake;
the brake module is arranged on a vehicle and used for avoiding forward collision to compensate the brake strength of the whole vehicle;
the GPS module is arranged on the vehicle and used for collecting the position information of the vehicle;
the main controller is arranged on a vehicle and used for collecting the vehicle speed collected by the speed sensor, the vehicle acceleration collected by the acceleration sensor, the vehicle position information collected by the GPS module and the distance between the vehicle and the front vehicle collected by the distance measuring sensor, and determining whether the early warning prompter and the brake module work or not according to the feedback result of the remote control terminal;
the wireless transmission module is arranged on a vehicle and used for transmitting vehicle information acquired by the main controller to the remote control terminal and transmitting feedback information from the remote control terminal to the main controller;
the remote control terminal is used for analyzing and receiving vehicle information, judging the position of a front vehicle according to the GPS position information, completing self-learning of the expected braking strength of a driver and judging whether a forward collision risk exists or not.
The speed sensor is selected to be a photoelectric sensor; the acceleration sensor is a piezoresistive acceleration sensor; the type of the distance measuring sensor is selected to be a laser radar; the brake module comprises a main brake disc brake and an auxiliary brake device hydraulic retarder; the main controller comprises an input loop, a microcontroller and an output loop; the early warning prompter is selected from a buzzer and a prompting lamp; the GPS module is selected as a combined inertial navigation system; the wireless transmission module carries out signal transmission through a 5G network; and the remote control terminal is selected as an intelligent networked automobile management platform server.
The following describes the embodiments of the present invention with reference to fig. 1 to 4:
step 1: the remote control terminal judges whether the self vehicle and the front vehicle keep a running state and have a forward collision danger or not according to the speed of the self vehicle, the acceleration of the self vehicle, the speed of the front vehicle, the acceleration of the front vehicle and the distance between the self vehicle and the front vehicle at the current moment;
in the step 1, the remote control terminal confirms a front vehicle according to the GPS information of the vehicle in the intelligent network connection;
in step 1, the speed of the bicycle is vrThe acceleration of the bicycle is arThe speed of the front vehicle is vfThe acceleration of the front vehicle is afThe distance between the bicycle and the front bicycle is D0；
The main controller receives v acquired by the speed sensorrAnd sending the data to a remote control terminal through a wireless transmission module;
the main controller receives a acquired by the acceleration sensorrAnd sending the data to a remote control terminal through a wireless transmission module;
d acquired by main controller receiving distance measuring sensor0And sending the data to a remote control terminal through a wireless transmission module;
in the step 1, the step of judging whether the self vehicle and the front vehicle keep the driving state or not has the risk of forward collision is as follows:
if ar>afThe speed of the bicycle is gradually faster than that of the front bicycle, so that the danger of forward collision exists;
if ar<afAnd v isr<vfThe distance between the bicycle and the front bicycle is enlarged, and the danger of forward collision is avoided;
if ar<afAnd v isr>vfIf D is>0, no forward collision danger exists, and if D is less than or equal to 0, the forward collision danger exists;
d is specifically calculated as follows:
Figure BDA0002115055340000101
wherein D is when vr＝vfThe distance between the current vehicle and the preceding vehicle;
step 2: if the forward collision risk exists, the remote control terminal expects the brake intensity a of the driver through the recurrent neural networkePredicting, wherein a cyclic neural network self-learning algorithm is shown in figure 3;
the brake intensity a expected by the driver in step 2eThe specific process for predicting comprises the following steps:
step 2.1: constructing a training set of a recurrent neural network;
step 2.1, constructing a training set of the recurrent neural network as follows:
the remote control terminal extracts relevant parameters of the braking time in the latest n times of forward risk avoiding process of the self-vehicle as a training set, wherein m is the extracted 1 st forward risk avoiding process, and the training set is as follows:
xi＝({v'r,i,v'f,i,a'r,i,a'f,i,D'0,i},0＜m≤i≤n+m,m,n,i∈Z)
wherein, v'r,iIs the speed v 'of the vehicle at the braking moment in the ith forward danger avoiding process'f,iIs the front vehicle speed a 'of the braking time in the ith forward danger avoiding process'r,iFor the ith forward risk avoidance processFrom vehicle acceleration at moment of motion, a'f,iIs the front vehicle acceleration D 'of the braking moment in the ith forward danger avoiding process'0,iThe distance between the self vehicle and the front vehicle at the braking moment in the ith forward danger avoiding process is obtained;
recording the braking intensity set y of the braking moment in the last n times of forward risk avoidance processi：
yi＝({ad,i}，m≤i≤n+m,m,n,i∈Z)
Wherein, ad,iThe real braking strength at the braking moment in the ith forward risk avoiding process is obtained;
step 2.2: training the cyclic neural network according to the training set to obtain a trained cyclic neural network;
the specific process of training the recurrent neural network according to the training set in step 2.2 is as follows:
determining a specific structure of a recurrent neural model, and constructing a recurrent neural network model; the constructed recurrent neural model uses a model with 1 input layer, 5 hidden layers and 1 output layer.
Model initialization: randomly initializing a weight matrix U, W, V and bias matrixes b and c in the model parameters; the hidden state of the recurrent neural network model at the braking moment in the ith forward risk avoidance process is recorded as hiAnd recording the predicted value of the model
Figure BDA0002115055340000111
Activation function f (x) is typically tanh, b is a bias in a linear relationship, and activation function g (x) is typically a Softmax function; a recurrent neural network can generally be written as follows:
Figure BDA0002115055340000112
Figure BDA0002115055340000113
forward propagation training: inputting training sample data into a recurrent neural model, and obtaining prediction of the recurrent neural model under initial model parameters through forward propagationValue, by continuously reducing the predicted value
Figure BDA0002115055340000114
With the true value yiAdjusting the model parameters by the difference value;
and (3) back propagation training: selecting a loss function of the model as an optimization target, and taking a model parameter weight matrix U, W, V and bias matrixes b and c as optimization objects; iterating the model parameters by using a gradient descent method according to the error;
a cross entropy function L oss was chosen as the loss function, denoted L, expressed as follows:
Figure BDA0002115055340000121
calculating the gradient of the weight matrix V and the bias matrix c:
Figure BDA0002115055340000122
Figure BDA0002115055340000123
computing i-time hidden state gradientiWhen the gradient loss of the i is determined by the gradient loss corresponding to the current forward collision risk avoidance and the gradient loss of the next forward collision risk avoidance i +1 in the reverse transmission, the gradient loss of the i is determined by the gradient loss corresponding to the current forward collision risk avoidance and the gradient loss of the next forward collision risk avoidance i +1, and the reference deep neural network is used for referencei+1Recursion to each otheriThe function diag represents taking the diagonal elements of the matrix:
Figure BDA0002115055340000124
calculating the gradients of the weight matrix W, U and the bias matrix b:
Figure BDA0002115055340000125
Figure BDA0002115055340000126
Figure BDA0002115055340000127
through repeated iteration:
step 2.3: determining a trained recurrent neural network model;
the specific process of training the recurrent neural network according to the training set in step 2.3 is as follows:
using the same training set xiInputting the optimized recurrent neural network model again;
combining forward propagation training and backward propagation training to compare predicted values
Figure BDA0002115055340000128
With the true value yiError between, predicted brake strength
Figure BDA0002115055340000129
With true braking strength yiThe average error should be less than a certain threshold. If the error meets the requirement, determining the model parameters; if the error does not meet the requirement, the step 2.2 is repeated to adjust the parameters until the error meets the requirement;
step 2.4: the remote control terminal sends the current speed v of the vehiclerSpeed v of front vehiclefAcceleration a of bicyclerAcceleration a of the front vehiclefDistance D between the bicycle and the front bicycle0Substituting the optimized recurrent neural network model to continuously predict the expected brake intensity of the drivere；
And step 3: the remote control terminal calculates the alarm time;
in the step 3, the calculation of the allowable risk avoidance operation time of the driver by the remote control terminal specifically comprises the following steps:
the current time is T0The driver's expected braking intensity is ae；
At the latest operation time T when the driver allows danger avoidanceeThe speed of the bicycle is ve,rThe speed of the front vehicle is ve,fThe distance D (t) between the two vehicles after the driver brakes at the latest operation time allowing danger avoidance is calculated according to the following formula:
ve,r＝(Te-T0)·ar+vr
ve,f＝(Te-T0)·af+vf
Figure BDA0002115055340000131
if the driver wants to ensure that the driver does not collide with the front vehicle, the latest operation time T for avoiding danger is allowed for the drivereD (T) is equal to or greater than 0, i.e., the discriminant Δ is equal to or greater than 0, and the critical condition is that Δ is equal to 0 and T is equal toe-T0For the driver's allowable safe-keeping operation time, d (t) the discriminant Δ is calculated as follows:
Figure BDA0002115055340000132
wherein, T0Is the current time, TeLatest operation moment v of driver's permission to avoid dangere,rVehicle speed v of the vehicle at the latest operating moment allowed for the driver to avoid dangere,fVehicle speed before the latest operation time allowed for driver to avoid danger, vrFor the current time the speed of the bicycle, arAcceleration of the vehicle at the present moment, vfFor the current time the speed of the bicycle, afAcceleration of the vehicle at the present moment, D0D (t) is a distance between the self vehicle and the front vehicle at the current moment, D (t) is a distance function between the two vehicles after the driver brakes at the latest operation moment allowing risk avoidance, and t is the time after the latest operation moment allowing risk avoidance brakes;
the latest operation time T allowed by the driver to avoid danger can be calculated through the formulae；
As shown in fig. 4, but in actual practice, the alarm time TaShould be earlier than the latest operation time T of the driver for avoiding dangereOn the one hand, the driver should be given an emergency response time t after the alarm is givendOn the other hand, the acceleration change of the vehicle is a continuous processA certain response time tvThen alarm time TaCan be expressed as follows:
Ta＝Te-td-tv
and 4, step 4: when there is a forward collision danger, the driver is at warning time TaBefore, danger avoiding operation is not carried out, the remote control terminal transmits danger signals to the early warning prompter through the wireless transmission module, and the main controller enables the early warning prompter to work and gives out an alarm.
And 5: if the driver carries out danger avoiding operation before the alarming moment, the remote control terminal further judges whether a forward collision danger exists;
in step 5, the remote control terminal further judges whether the danger of forward collision exists:
the remote control terminal brakes the real braking time T of the driverdWith true braking strength adCalculating the distance D (t) between two vehicles after the driver brakes at the latest operation time allowing risk avoidance:
ve,r＝(Te-Td)·a'r+v'r
ve,f＝(Te-Td)·a'f+v'f
Figure BDA0002115055340000141
the remote control terminal brakes according to the real braking time T of the driverdWith true braking strength adCalculating D (t) discriminant Δ:
Figure BDA0002115055340000142
wherein, TdFor the actual braking moment, TeLatest operation moment v of driver's permission to avoid dangere,rVehicle speed v of the vehicle at the latest operating moment allowed for the driver to avoid dangere,fVehicle speed v 'ahead of time of latest operation allowed for driver to avoid danger'rIs the speed of the vehicle at the moment of real braking of the driver, a'rFor the driverReal braking moment self-vehicle acceleration v'fIs the speed of the vehicle at the moment of real braking of the driver, a'fIs the front vehicle acceleration of the driver at the real braking moment D'0The distance between the self vehicle and the front vehicle at the real braking moment of the driver, D (t) is a function of the distance between the two vehicles after the driver brakes at the latest operation moment allowing danger avoidance, and t is the time after the driver brakes at the latest operation moment allowing danger avoidance;
if delta is less than 0, a forward collision danger exists, the remote control terminal transmits a danger signal to the main controller through the wireless transmission module, and the main controller enables the early warning controller to work and gives an alarm;
step 6: calculating the compensation braking intensity;
in step 6, the compensation braking intensity is calculated as:
when the alarm is given, the driver reacts at the emergency reaction time tdShould be operated accordingly, at the vehicle response time TvWhen the danger avoiding operation is completed, the remote control terminal transmits a danger removing signal to the main controller through the wireless transmission module, and the main controller stops the early warning controller to stop working and remove the alarm; otherwise, the remote control terminal calculates whether the braking intensity needs to be compensated or not through the safety distance; at the vehicle response time TvThe remote control terminal extracts the speed v of the vehiclev,rSpeed v of the preceding vehiclev,fAcceleration a of bicyclev,rAcceleration a of the preceding vehiclev,fDistance D between the bicycle and the preceding bicyclevWhen v isv,r＝vv,fAt a distance of D1：
Figure BDA0002115055340000151
Tv＝Te-tv
When D is present1When the alarm is more than or equal to 0, the remote control terminal transmits the danger removing signal to the main controller through the wireless transmission module, and the main controller enables the early warning controller to stop working and remove the alarm. When D is present1<When 0, the remote control terminal transmits information to enable the main controller to command the brake module to perform brake intensity compensation, and the compensation value is acAfter compensation control, D should be satisfied1≥0：
Figure BDA0002115055340000152
From the above formula, a minimum value a of the compensated braking intensity can be derivedcminComprises the following steps:
Figure BDA0002115055340000153
wherein, TvAs the vehicle response time, TeThe latest operation time t allowed by the driver to avoid dangerdDriver emergency response time, vv,rVehicle speed, v, of the vehicle at the time of responsev,fThe vehicle speed before the time of response of the vehicle, av,rAcceleration of the vehicle from the moment of response of the vehicle, av,fAcceleration of the vehicle ahead of the time of response of the vehicle, acTo compensate for braking strength, DvDistance between the vehicle and the preceding vehicle at the time of response of the vehicle, D1Is when v isv,r＝vv,fThe distance between the current vehicle and the front vehicle;
and 7: when the braking intensity compensation is needed, the remote control terminal transmits the minimum compensation braking intensity to the main controller through the wireless transmission module, the main controller sends a command to the braking module, and the braking module completes the compensation of the braking intensity through the regulation and control of the main brake and the auxiliary braking device.
It should be understood that parts of the specification not set forth in detail are well within the prior art.
It should be understood that the above-mentioned preferred embodiments are described in some detail, and not intended to limit the scope of the invention, and those skilled in the art will be able to make alterations and modifications without departing from the spirit and scope of the invention as defined by the appended claims.

Claims (4)

1. A car forward collision early warning control method of a recurrent neural network is characterized by comprising the following steps:
the system comprises a speed sensor, an acceleration sensor, a distance measuring sensor, a main controller, a brake module, an early warning prompter, a GPS module, a wireless transmission module and a remote control terminal;
the main controller is respectively connected with the speed sensor, the acceleration sensor, the distance measuring sensor, the braking module, the early warning prompter and the wireless transmission module in sequence through leads; the wireless transmission module is connected with the remote control terminal in a wireless communication mode;
the speed sensor is arranged on the vehicle and used for acquiring the speed of the vehicle;
the acceleration sensor is arranged on the vehicle and used for acquiring the acceleration of the vehicle;
the distance measuring sensor is arranged at the front end of the vehicle and is used for collecting the distance between the self vehicle and the front vehicle;
the early warning prompter is arranged near a driver on the vehicle, comprises an early warning indicator lamp and a buzzer and is used for prompting the driver to brake;
the brake module is arranged on a vehicle and used for avoiding forward collision to compensate the brake strength of the whole vehicle;
the GPS module is arranged on the vehicle and used for collecting the position information of the vehicle;
the main controller is arranged on a vehicle and used for collecting the vehicle speed collected by the speed sensor, the vehicle acceleration collected by the acceleration sensor, the vehicle position information collected by the GPS module and the distance between the vehicle and the front vehicle collected by the distance measuring sensor, and determining whether the early warning prompter and the brake module work or not according to the feedback result of the remote control terminal;
the wireless transmission module is arranged on a vehicle and used for transmitting vehicle information acquired by the main controller to the remote control terminal and transmitting feedback information from the remote control terminal to the main controller;
the remote control terminal is used for analyzing and receiving vehicle information, judging the position of a front vehicle according to the GPS position information, completing self-learning of the expected braking strength of a driver and judging whether a forward collision risk exists or not;
the automobile forward collision early warning control method of the recurrent neural network comprises the following steps:
step 1: the remote control terminal judges whether the self vehicle and the front vehicle keep a running state and have a forward collision danger or not according to the speed of the self vehicle, the acceleration of the self vehicle, the speed of the front vehicle, the acceleration of the front vehicle and the distance between the self vehicle and the front vehicle at the current moment;
step 2: if the forward collision risk exists, the remote control terminal expects the brake intensity a of the driver through the recurrent neural networkeCarrying out prediction;
and step 3: the remote control terminal calculates the alarm time;
and 4, step 4: when there is a forward collision danger, the driver is at warning time TaBefore the danger avoiding operation is carried out, the remote control terminal transmits danger signals to the early warning prompter through the wireless transmission module, and the main controller enables the early warning prompter to work and gives an alarm;
and 5: if the driver carries out danger avoiding operation before the alarming moment, the remote control terminal further judges whether a forward collision danger exists;
step 6: calculating the compensation braking intensity;
and 7: when the braking intensity compensation is needed, the remote control terminal transmits the minimum compensation braking intensity to the main controller through the wireless transmission module, the main controller sends a command to the braking module, and the braking module completes the compensation of the braking intensity through regulating and controlling the main brake and the auxiliary braking device;
in the step 1, the remote control terminal confirms a front vehicle according to the GPS information of the vehicle in the intelligent network connection;
in step 1, the speed of the bicycle is vrThe acceleration of the bicycle is arThe speed of the front vehicle is vfThe acceleration of the front vehicle is afThe distance between the bicycle and the front bicycle is D0；
The main controller receives v acquired by the speed sensorrAnd sending the data to a remote control terminal through a wireless transmission module;
the main controller receives a acquired by the acceleration sensorrAnd sending the data to a remote control terminal through a wireless transmission module;
d acquired by main controller receiving distance measuring sensor0And sending the data to a remote control terminal through a wireless transmission module;
in the step 1, the step of judging whether the self vehicle and the front vehicle keep the driving state or not has the risk of forward collision is as follows:
if ar>afThe speed of the bicycle is gradually faster than that of the front bicycle, so that the danger of forward collision exists;
if ar<afAnd v isr<vfThe distance between the bicycle and the front bicycle is enlarged, and the danger of forward collision is avoided;
if ar<afAnd v isr>vfIf D is>0, no forward collision danger exists, and if D is less than or equal to 0, the forward collision danger exists;
d is specifically calculated as follows:
Figure FDA0002512944070000021
wherein D is when vr＝vfThe distance between the current vehicle and the preceding vehicle;
the brake intensity a expected by the driver in step 2eThe specific process for predicting comprises the following steps:
step 2.1: constructing a training set of a recurrent neural network;
step 2.1, constructing a training set of the recurrent neural network as follows:
the remote control terminal extracts relevant parameters of the braking time in the latest n times of forward risk avoiding process of the self-vehicle as a training set, wherein m is the extracted 1 st forward risk avoiding process, and the training set is as follows:
xi＝({v'r,i,v'f,i,a'r,i,a'f,i,D'0,i},0＜m≤i≤n+m,m,n,i∈Z)
wherein, v'r,iIs the speed v 'of the vehicle at the braking moment in the ith forward danger avoiding process'f,iIs the front vehicle speed a 'of the braking time in the ith forward danger avoiding process'r,iIs the self vehicle acceleration at the braking moment in the ith forward danger avoiding process'f,iIs the front vehicle acceleration D 'of the braking moment in the ith forward danger avoiding process'0,iThe distance between the self vehicle and the front vehicle at the braking moment in the ith forward danger avoiding process is obtained;
recording the braking intensity set y of the braking moment in the last n times of forward risk avoidance processi：
yi＝({ad,i}，m≤i≤n+m,m,n,i∈Z)
Wherein, ad,iThe real braking strength at the braking moment in the ith forward risk avoiding process is obtained;
step 2.2: training the cyclic neural network according to the training set to obtain a trained cyclic neural network;
the specific process of training the recurrent neural network according to the training set in step 2.2 is as follows:
determining a specific structure of a recurrent neural model, and constructing a recurrent neural network model; the constructed recurrent neural model adopts a model with 1 input layer, 5 hidden layers and 1 output layer;
model initialization: randomly initializing a weight matrix U, W, V and bias matrixes b and c in the model parameters; the hidden state of the recurrent neural network model at the braking moment in the ith forward risk avoidance process is recorded as hiAnd recording the predicted value of the model
Figure FDA0002512944070000031
Activation function f (x) is typically tanh, b is a bias in a linear relationship, and activation function g (x) is typically a Softmax function; a recurrent neural network can generally be written as follows:
Figure FDA0002512944070000032
Figure FDA0002512944070000033
forward directionAnd (3) propagation training: inputting training sample data into a recurrent neural model, obtaining a predicted value of the recurrent neural model under initial model parameters through forward propagation, and continuously reducing the predicted value
Figure FDA0002512944070000034
With the true value yiAdjusting the model parameters by the difference value;
and (3) back propagation training: selecting a loss function of the model as an optimization target, and taking a model parameter weight matrix U, W, V and bias matrixes b and c as optimization objects; iterating the model parameters by using a gradient descent method according to the error;
a cross entropy function L oss was chosen as the loss function, denoted L, expressed as follows:
Figure FDA0002512944070000041
calculating the gradient of the weight matrix V and the bias matrix c:
Figure FDA0002512944070000042
Figure FDA0002512944070000043
computing i-time hidden state gradientiWhen the gradient loss of the i is determined by the gradient loss corresponding to the current forward collision risk avoidance and the gradient loss of the next forward collision risk avoidance i +1 in the reverse transmission, the gradient loss of the i is determined by the gradient loss corresponding to the current forward collision risk avoidance and the gradient loss of the next forward collision risk avoidance i +1, and the reference deep neural network is used for referencei+1Recursion to each otheriThe function diag represents taking the diagonal elements of the matrix:
Figure FDA0002512944070000044
calculating the gradients of the weight matrix W, U and the bias matrix b:
Figure FDA0002512944070000045
Figure FDA0002512944070000046
Figure FDA0002512944070000047
through repeated iteration:
step 2.3: determining a trained recurrent neural network model;
the specific process of training the recurrent neural network according to the training set in step 2.3 is as follows:
using the same training set xiInputting the optimized recurrent neural network model again;
combining forward propagation training and backward propagation training to compare predicted values
Figure FDA0002512944070000048
With the true value yiError between, predicted brake strength
Figure FDA0002512944070000049
With true braking strength yiThe average error should be less than a certain threshold; if the error meets the requirement, determining the model parameters; if the error does not meet the requirement, the step 2.2 is repeated to adjust the parameters until the error meets the requirement;
step 2.4: the remote control terminal sends the current speed v of the vehiclerSpeed v of front vehiclefAcceleration a of bicyclerAcceleration a of the front vehiclefDistance D between the bicycle and the front bicycle0Substituting the optimized recurrent neural network model to continuously predict the expected brake intensity of the drivere。
2. The automobile forward collision early warning control method of the recurrent neural network as claimed in claim 1, wherein the step 3 of calculating the allowable risk avoidance operation time of the driver by the remote control terminal specifically comprises:
the current time is T0The driver's expected braking intensity is ae；
At the latest operation time T when the driver allows danger avoidanceeThe speed of the bicycle is ve,rThe speed of the front vehicle is ve,fThe distance D (t) between the two vehicles after the driver brakes at the latest operation time allowing danger avoidance is calculated according to the following formula:
ve,r＝(Te-T0)·ar+vr
ve,f＝(Te-T0)·af+vf
Figure FDA0002512944070000051
if the driver wants to ensure that the driver does not collide with the front vehicle, the latest operation time T for avoiding danger is allowed for the drivereD (T) is equal to or greater than 0, i.e., the discriminant Δ is equal to or greater than 0, and the critical condition is that Δ is equal to 0 and T is equal toe-T0For the driver's allowable safe-keeping operation time, d (t) the discriminant Δ is calculated as follows:
Figure FDA0002512944070000052
wherein, T0Is the current time, TeLatest operation moment v of driver's permission to avoid dangere,rVehicle speed v of the vehicle at the latest operating moment allowed for the driver to avoid dangere,fVehicle speed before the latest operation time allowed for driver to avoid danger, vrFor the current time the speed of the bicycle, arAcceleration of the vehicle at the present moment, vfFor the current time the speed of the bicycle, afAcceleration of the vehicle at the present moment, D0D (t) is a distance between the self vehicle and the front vehicle at the current moment, D (t) is a distance function between the two vehicles after the driver brakes at the latest operation moment allowing risk avoidance, and t is the time after the latest operation moment allowing risk avoidance brakes;
the latest operation time T allowed by the driver to avoid danger can be calculated through the formulae；
However, in practical cases, the alarm time TaShould be earlier than the latest operation time T of the driver for avoiding dangereOn the one hand, the driver should be given an emergency response time t after the alarm is givendOn the other hand, the acceleration change of the automobile is a continuous process and needs a certain response time tvThen alarm time TaCan be expressed as follows:
Ta＝Te-td-tv。
3. the automobile forward collision early warning control method of the recurrent neural network as claimed in claim 1, wherein in step 5, the remote control terminal further determines whether there is a forward collision risk as:
the remote control terminal brakes the real braking time T of the driverdWith true braking strength adCalculating the distance D (t) between two vehicles after the driver brakes at the latest operation time allowing risk avoidance:
ve,r＝(Te-Td)·a'r+v'r
ve,f＝(Te-Td)·a'f+v'f
Figure FDA0002512944070000061
the remote control terminal brakes according to the real braking time T of the driverdWith true braking strength adCalculating D (t) discriminant Δ:
Figure FDA0002512944070000062
wherein, TdFor the actual braking moment, TeLatest operation moment v of driver's permission to avoid dangere,rVehicle speed v of the vehicle at the latest operating moment allowed for the driver to avoid dangere,fVehicle speed v 'ahead of time of latest operation allowed for driver to avoid danger'rFor the driver at the actual braking momentFast a'rIs the acceleration v 'of the driver at the real braking moment'fIs the speed of the vehicle at the moment of real braking of the driver, a'fIs the front vehicle acceleration of the driver at the real braking moment D'0The distance between the self vehicle and the front vehicle at the real braking moment of the driver, D (t) is a function of the distance between the two vehicles after the driver brakes at the latest operation moment allowing danger avoidance, and t is the time after the driver brakes at the latest operation moment allowing danger avoidance;
if delta <0, a forward collision danger exists, the remote control terminal transmits a danger signal to the main controller through the wireless transmission module, and the main controller enables the early warning controller to work and send out an alarm.
4. The automobile forward collision early warning control method of the recurrent neural network as claimed in claim 1, wherein said compensation braking intensity in step 6 is calculated as:
when the alarm is given, the driver reacts at the emergency reaction time tdShould be operated accordingly, at the vehicle response time TvWhen the danger avoiding operation is completed, the remote control terminal transmits a danger removing signal to the main controller through the wireless transmission module, and the main controller stops the early warning controller to stop working and remove the alarm; otherwise, the remote control terminal calculates whether the braking intensity needs to be compensated or not through the safety distance; at the vehicle response time TvThe remote control terminal extracts the speed v of the vehiclev,rSpeed v of the preceding vehiclev,fAcceleration a of bicyclev,rAcceleration a of the preceding vehiclev,fDistance D between the bicycle and the preceding bicyclevWhen v isv,r＝vv,fAt a distance of D1：
Figure FDA0002512944070000071
Tv＝Te-tv
When D is present1When the alarm is more than or equal to 0, the remote control terminal transmits a danger removing signal to the main controller through the wireless transmission module, and the main controller enables the early warning controller to stop working and remove the alarm; when D is present1<When 0, the remote control terminal transmits information to enable the main controller to command the brake module to perform brake intensity compensation, and the compensation value is acAfter compensation control, D should be satisfied1≥0：
Figure FDA0002512944070000072
From the above formula, a minimum value a of the compensated braking intensity can be derivedcminComprises the following steps:
Figure FDA0002512944070000073
wherein, TvAs the vehicle response time, TeThe latest operation time t allowed by the driver to avoid dangerdDriver emergency response time, vv,rVehicle speed, v, of the vehicle at the time of responsev,fThe vehicle speed before the time of response of the vehicle, av,rAcceleration of the vehicle from the moment of response of the vehicle, av,fAcceleration of the vehicle ahead of the time of response of the vehicle, acTo compensate for braking strength, DvDistance between the vehicle and the preceding vehicle at the time of response of the vehicle, D1Is when v isv,r＝vv,fThe distance between the bicycle and the front bicycle.
CN201910587792.XA 2019-07-02 2019-07-02 Automobile forward collision early warning control system and method of cyclic neural network Active CN110329249B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910587792.XA CN110329249B (en) 2019-07-02 2019-07-02 Automobile forward collision early warning control system and method of cyclic neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910587792.XA CN110329249B (en) 2019-07-02 2019-07-02 Automobile forward collision early warning control system and method of cyclic neural network

Publications (2)

Publication Number Publication Date
CN110329249A CN110329249A (en) 2019-10-15
CN110329249B true CN110329249B (en) 2020-08-07

Family

ID=68143859

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910587792.XA Active CN110329249B (en) 2019-07-02 2019-07-02 Automobile forward collision early warning control system and method of cyclic neural network

Country Status (1)

Country Link
CN (1) CN110329249B (en)

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111090747A (en) * 2019-12-03 2020-05-01 国家电网有限公司 Power communication fault emergency disposal method based on neural network classification
CN111081067B (en) * 2019-12-27 2021-07-20 武汉大学 Vehicle collision early warning system and method based on IGA-BP neural network under vehicle networking environment
CN113092127A (en) * 2021-03-11 2021-07-09 东风柳州汽车有限公司 Test method, device, storage medium and device for front collision early warning
WO2022193137A1 (en) * 2021-03-16 2022-09-22 华为技术有限公司 Vehicle control method and device
CN113516862A (en) * 2021-07-22 2021-10-19 中国第一汽车股份有限公司 Early warning method and device, electronic equipment and storage medium
CN113920780B (en) * 2021-09-01 2022-12-16 同济大学 Cloud and mist collaborative personalized forward collision risk early warning method based on federal learning
CN113895453A (en) * 2021-10-22 2022-01-07 青岛海尔智能技术研发有限公司 Vehicle control method, device, vehicle and storage medium
US11851068B2 (en) 2021-10-25 2023-12-26 Ford Global Technologies, Llc Enhanced target detection
CN114454876B (en) * 2022-04-13 2022-07-26 所托(杭州)汽车智能设备有限公司 Vehicle collision determination method, detection device, and storage medium

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102109821B (en) * 2010-12-30 2013-08-07 中国科学院自动化研究所 System and method for controlling adaptive cruise of vehicles
CN102745194B (en) * 2012-06-19 2015-02-25 东南大学 Self-adaption alarming method for preventing tailgating with front car on expressway
CN104691535A (en) * 2013-12-09 2015-06-10 青岛盛嘉信息科技有限公司 Brake assist system
JP6237580B2 (en) * 2014-11-13 2017-11-29 株式会社デンソー Motor control device
CN104494550B (en) * 2014-12-15 2017-09-26 江苏大学 A kind of computational methods for vehicle active collision avoidance distance
CN105137970B (en) * 2015-07-31 2018-03-16 奇瑞汽车股份有限公司 Vehicle obstacle-avoidance method and device
CN105151024A (en) * 2015-08-20 2015-12-16 奇瑞汽车股份有限公司 Vehicle brake control method and device
CN105835854B (en) * 2016-03-17 2018-11-16 奇瑞汽车股份有限公司 A kind of emergency braking control system and its control method
CN107697045A (en) * 2017-09-07 2018-02-16 厦门大学 A kind of pilotless automobile automatic brake controller and method
CN109591811B (en) * 2017-09-28 2020-08-14 华为技术有限公司 Vehicle braking method, device and storage medium

Also Published As

Publication number Publication date
CN110329249A (en) 2019-10-15

Similar Documents

Publication Publication Date Title
CN110329249B (en) Automobile forward collision early warning control system and method of cyclic neural network
EP3156298B1 (en) Driving aid arrangement, a vehicle and a method of controlling a longitudinal velocity of a vehicle
US11104333B2 (en) Emergency braking system, emergency braking method and semitrailer
CN109910879B (en) Vehicle safety anti-collision control method combining safe distance and collision time
EP2694344B1 (en) Estimation of road inclination
CN108428343A (en) A kind of more vehicle driving behavior analysis and danger early warning method and system
CN107264527B (en) Control method and device for preventing intelligent vehicle from being parked
KR101509693B1 (en) System and method of determining short term driving tendency of driver
CN111201173B (en) Method for determining a friction value of a contact between a vehicle tire and a lane and method for controlling a vehicle function of a vehicle
EP3741638A1 (en) Vehicle control device
CN110435647B (en) TTC vehicle safety anti-collision control method based on rolling optimization parameters
EP3666612A1 (en) Vehicle control device
CN110588623A (en) Large automobile safe driving method and system based on neural network
CN114148321B (en) Automatic emergency braking and safety evaluation method considering collision risk
CN108327691A (en) Obtain method of the long-term brake assemblies than change in torque
CN114222987A (en) Method for quantifying vehicle path following performance
CN112298136A (en) Driving control method, device and equipment for automatic driving vehicle and readable medium
CN113165615A (en) Vehicle control method and device
JP2012059058A (en) Risk estimation device and program
EP2862157B1 (en) System and method pertaining to vehicle trains
CN114734966A (en) Automatic emergency braking system and method based on camera and cloud real-time map
EP4017773B1 (en) A method for quantifying extreme traffic behavior
CN111856510A (en) Vehicle front collision prediction method based on laser radar
CN111653122A (en) Vehicle cooperative collision early warning system and control method thereof
US20230242129A1 (en) System, server, device and method for training and using neural networks for prediction of vehicle braking distance

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant